For one very simple problem, our SW engineering group talked about 4
very different ways to set up the problem, and no one in the group
could see any obvious reason to select one over the other. As
researchers we tend to think: just try one and if it doesn't work, try
a different one. Applied folks need to expect to do some of that as
well, but they would like to have at least some guidance in addition.

This all seems very obvious to those of us who've done a lot of GP. We
tend to just start with +,-,%,*,X, and rand(), and then maybe think
about whether we need if's and conditional expressions, and maybe
loops. But this looks very strange and limited to a software
engineer.

I couldn't think of much in the way of general principles to help them
out, I could only point out the pluses and minuses of any specific
proposal. I also talked about how GPPS showed that a simple universal
primitive set could be applied very broadly (with known performance
drawbacks).

Forrest